Skip to main content
Planck sits between your application and supported AI providers. It gives your team one place to call models, control production behavior, inspect every request, and improve prompts without rebuilding the same infrastructure for each provider. You can route model traffic through Planck or keep your existing provider path and add Planck only for observability.

One model interface

Use OpenAI-compatible or Anthropic-compatible APIs across supported text, reasoning, embedding, reranking, image, audio, and video models.

Production controls

Apply provider routing, fallbacks, caching, rate limits, and prompt security at the gateway.

Request-level observability

Investigate inputs, outputs, cost, latency, users, sessions, scores, feedback, and errors in context.

Prompts outside your code

Version, assemble, test, and deploy prompts without tying every change to an application release.

​
From application code to a production record

With the gateway path, your application sends a familiar model request and Planck handles the work around it.
  1. Planck authenticates the request and checks its configuration.
  2. It selects an eligible provider endpoint and applies supported controls.
  3. The provider response is normalized for the API style you chose.
  4. Planck records the request, response, usage, cost, timing, and application context for investigation and analysis.

​
The gateway change is small

If you already use the OpenAI SDK, the basic gateway integration changes the client configuration—not the shape of every model call.
Use a Planck API key with either Planck-managed provider access or provider keys configured in your organization.

​
Choose the integration that fits your application

Route requests through Planck

Recommended for most new integrations. Use unified model access, routing, billing, controls, and automatic observability.

Keep calling providers directly

Add automatic logging while model traffic continues on its existing provider path.

Control exactly what is recorded

Send completed model request records explicitly with a logger SDK or REST.

Trace the whole workflow

Connect model requests with tool calls, retrieval, and application-specific operations.
These paths can be combined. For example, route model calls through the gateway and add custom traces for the tool and retrieval steps around them. See Choose an Integration for a side-by-side comparison.

​
Where to go next

Make your first request

Send a copy-paste example and verify the resulting production record.

Explore every feature

Scan the gateway, reliability, observability, prompt, and integration map.

Find the right model

Discover supported operations, modalities, parameters, and provider endpoints.

Build an application

Follow end-to-end guides for model calls, streaming, prompting, and common production patterns.